Trung Dinh

> Also, I realized that there is no metrics to indicate if a `manual_compaction` job is running I looked at the code. Currently, `compaction` and `"manual" compaction` are mixed together....

Btw, I guess this code path can be parrallelized https://github.com/neondatabase/neon/blob/28e882a80f2bc48355ec4eac403746127382979b/pageserver/src/tenant/timeline/compaction.rs#L3126?

> As you might have known we only have a single compaction thread for a tenant Yup! That's another observation. I am just curious why we didn't pursue it? My...

Our model (self-hosted) is that we either have 1 (very) large tenant (>50TB) per pageserver or many small tenants per pageserver. I wonder if this is something we can collaborate...

> (we are tracking this internally at https://databricks.atlassian.net/browse/LKB-673) Let us know if anything we can help with. We have resource to seriously build a strong collaboration.

Just curious, what's the motivation behind exposing a grpc server from pageserver? Seems like a large initiative. How does the community help with this?

For my understanding, after this initiative, compute nodes will use gRPC to talk to pageserver instead of using the `COPY` command in postgres?

Thank you for the clarification! the communicator process is also a WIP? We would extract part of the current pageserver code to be that communicator process?